blob: cb89f2f64fdc7ce88870a630968073864a39d318 [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * Health-checks functions.
3 *
Willy Tarreau26c25062009-03-08 09:38:41 +01004 * Copyright 2000-2009 Willy Tarreau <w@1wt.eu>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02005 * Copyright 2007-2009 Krzysztof Piotr Oledzki <ole@ans.pl>
Willy Tarreaubaaee002006-06-26 02:48:02 +02006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 */
13
Willy Tarreaub8816082008-01-18 12:18:15 +010014#include <assert.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020015#include <ctype.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020016#include <errno.h>
17#include <fcntl.h>
Willy Tarreau9b39dc52014-07-08 00:54:10 +020018#include <signal.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090019#include <stdarg.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020020#include <stdio.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020021#include <stdlib.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020022#include <string.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020023#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020024#include <unistd.h>
25#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040026#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090027#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020028#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020029#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020030#include <arpa/inet.h>
31
Willy Tarreauc7e42382012-08-24 19:22:53 +020032#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020033#include <common/compat.h>
34#include <common/config.h>
35#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020036#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020037#include <common/time.h>
Christopher Fauletcfda8472017-10-20 15:40:23 +020038#include <common/hathreads.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020039
40#include <types/global.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020041#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010042#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020043
44#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020045#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010046#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020047#include <proto/fd.h>
48#include <proto/log.h>
Willy Tarreau53a47662017-08-28 10:53:00 +020049#include <proto/mux_pt.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020050#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020051#include <proto/port_range.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010052#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010053#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010054#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020055#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020056#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010057#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020058#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020059#include <proto/log.h>
60#include <proto/dns.h>
61#include <proto/proto_udp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020062
Olivier Houchard9130a962017-10-17 17:33:43 +020063#ifdef USE_OPENSSL
64#include <proto/ssl_sock.h>
65#endif /* USE_OPENSSL */
66
Willy Tarreaubd741542010-03-16 18:46:54 +010067static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090068static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020069static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020070static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020071static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020072static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020073static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010074
Willy Tarreau8ceae722018-11-26 11:58:30 +010075DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert));
76DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule));
Christopher Faulet31dff9b2017-10-23 15:45:20 +020077
78
Simon Horman63a4a822012-03-19 07:24:41 +090079static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010080 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
81 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020082 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020083
Willy Tarreau23964182014-05-20 20:56:30 +020084 /* Below we have finished checks */
85 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020089
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010090 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
91 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
92 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020093
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010094 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
95 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
96 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020097
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010098 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
99 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200101 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200102
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100103 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
104 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
105 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900106
107 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
108 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200109 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200110};
111
Cyril Bontéac92a062014-12-27 22:28:38 +0100112const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
113 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
122 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
123 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
124};
125
Simon Horman63a4a822012-03-19 07:24:41 +0900126static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100127 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
128
129 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
130 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
131
132 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
133 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
134 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
135 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
136
137 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
138 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
139 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
140};
141
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200142/*
143 * Convert check_status code to description
144 */
145const char *get_check_status_description(short check_status) {
146
147 const char *desc;
148
149 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200150 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200151 else
152 desc = NULL;
153
154 if (desc && *desc)
155 return desc;
156 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200157 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200158}
159
160/*
161 * Convert check_status code to short info
162 */
163const char *get_check_status_info(short check_status) {
164
165 const char *info;
166
167 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200168 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200169 else
170 info = NULL;
171
172 if (info && *info)
173 return info;
174 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200175 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200176}
177
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100178const char *get_analyze_status(short analyze_status) {
179
180 const char *desc;
181
182 if (analyze_status < HANA_STATUS_SIZE)
183 desc = analyze_statuses[analyze_status].desc;
184 else
185 desc = NULL;
186
187 if (desc && *desc)
188 return desc;
189 else
190 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
191}
192
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200193/*
Simon Horman4a741432013-02-23 15:35:38 +0900194 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200195 * an adequate CHK_RES_* value. The new check->health is computed based
196 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200197 *
198 * Show information in logs about failed health check if server is UP
199 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200200 */
Simon Horman4a741432013-02-23 15:35:38 +0900201static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100202{
Simon Horman4a741432013-02-23 15:35:38 +0900203 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200204 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200205 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900206
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200207 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100208 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900209 check->desc[0] = '\0';
210 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200211 return;
212 }
213
Simon Horman4a741432013-02-23 15:35:38 +0900214 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200215 return;
216
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200217 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900218 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
219 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200220 } else
Simon Horman4a741432013-02-23 15:35:38 +0900221 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200222
Simon Horman4a741432013-02-23 15:35:38 +0900223 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200224 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900225 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200226
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100227 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900228 check->duration = -1;
229 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200230 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900231 check->duration = tv_ms_elapsed(&check->start, &now);
232 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200233 }
234
Willy Tarreau23964182014-05-20 20:56:30 +0200235 /* no change is expected if no state change occurred */
236 if (check->result == CHK_RES_NEUTRAL)
237 return;
238
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200239 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200240
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200241 switch (check->result) {
242 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200243 /* Failure to connect to the agent as a secondary check should not
244 * cause the server to be marked down.
245 */
246 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900247 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200248 (check->health > 0)) {
Christopher Faulet29f77e82017-06-08 14:04:45 +0200249 HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200250 report = 1;
251 check->health--;
252 if (check->health < check->rise)
253 check->health = 0;
254 }
255 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200256
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200257 case CHK_RES_PASSED:
258 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
259 if ((check->health < check->rise + check->fall - 1) &&
260 (check->result == CHK_RES_PASSED || check->health > 0)) {
261 report = 1;
262 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200263
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200264 if (check->health >= check->rise)
265 check->health = check->rise + check->fall - 1; /* OK now */
266 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200267
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200268 /* clear consecutive_errors if observing is enabled */
269 if (s->onerror)
270 s->consecutive_errors = 0;
271 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100272
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200273 default:
274 break;
275 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200276
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200277 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
278 (status != prev_status || report)) {
279 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200280 "%s check for %sserver %s/%s %s%s",
281 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200282 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100283 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100284 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200285 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200286
Emeric Brun5a133512017-10-19 14:42:30 +0200287 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200288
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100289 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200290 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
291 (check->health >= check->rise) ? check->fall : check->rise,
292 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200293
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200294 ha_warning("%s.\n", trash.area);
295 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
296 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200297 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200298}
299
Willy Tarreau4eec5472014-05-20 22:32:27 +0200300/* Marks the check <check>'s server down if the current check is already failed
301 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200302 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200303static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200304{
Simon Horman4a741432013-02-23 15:35:38 +0900305 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900306
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200307 /* The agent secondary check should only cause a server to be marked
308 * as down if check->status is HCHK_STATUS_L7STS, which indicates
309 * that the agent returned "fail", "stopped" or "down".
310 * The implication here is that failure to connect to the agent
311 * as a secondary check should not cause the server to be marked
312 * down. */
313 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
314 return;
315
Willy Tarreau4eec5472014-05-20 22:32:27 +0200316 if (check->health > 0)
317 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100318
Willy Tarreau4eec5472014-05-20 22:32:27 +0200319 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200320 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200321}
322
Willy Tarreauaf549582014-05-16 17:37:50 +0200323/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200324 * it isn't in maintenance, it is not tracking a down server and other checks
325 * comply. The rule is simple : by default, a server is up, unless any of the
326 * following conditions is true :
327 * - health check failed (check->health < rise)
328 * - agent check failed (agent->health < rise)
329 * - the server tracks a down server (track && track->state == STOPPED)
330 * Note that if the server has a slowstart, it will switch to STARTING instead
331 * of RUNNING. Also, only the health checks support the nolb mode, so the
332 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200333 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200334static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200335{
Simon Horman4a741432013-02-23 15:35:38 +0900336 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100337
Emeric Brun52a91d32017-08-31 14:41:55 +0200338 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200339 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100340
Emeric Brun52a91d32017-08-31 14:41:55 +0200341 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200342 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100343
Willy Tarreau3e048382014-05-21 10:30:54 +0200344 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
345 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100346
Willy Tarreau3e048382014-05-21 10:30:54 +0200347 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
348 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200349
Emeric Brun52a91d32017-08-31 14:41:55 +0200350 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200351 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100352
Emeric Brun5a133512017-10-19 14:42:30 +0200353 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100354}
355
Willy Tarreaudb58b792014-05-21 13:57:23 +0200356/* Marks the check <check> as valid and tries to set its server into stopping mode
357 * if it was running or starting, and provided it isn't in maintenance and other
358 * checks comply. The conditions for the server to be marked in stopping mode are
359 * the same as for it to be turned up. Also, only the health checks support the
360 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200361 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200362static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200363{
Simon Horman4a741432013-02-23 15:35:38 +0900364 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100365
Emeric Brun52a91d32017-08-31 14:41:55 +0200366 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200367 return;
368
Willy Tarreaudb58b792014-05-21 13:57:23 +0200369 if (check->state & CHK_ST_AGENT)
370 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100371
Emeric Brun52a91d32017-08-31 14:41:55 +0200372 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200373 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100374
Willy Tarreaudb58b792014-05-21 13:57:23 +0200375 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
376 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100377
Willy Tarreaudb58b792014-05-21 13:57:23 +0200378 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
379 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100380
Willy Tarreaub26881a2017-12-23 11:16:49 +0100381 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100382}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200383
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100384/* note: use health_adjust() only, which first checks that the observe mode is
385 * enabled.
386 */
387void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100388{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100389 int failed;
390 int expire;
391
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100392 if (s->observe >= HANA_OBS_SIZE)
393 return;
394
Willy Tarreaubb956662013-01-24 00:37:39 +0100395 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100396 return;
397
398 switch (analyze_statuses[status].lr[s->observe - 1]) {
399 case 1:
400 failed = 1;
401 break;
402
403 case 2:
404 failed = 0;
405 break;
406
407 default:
408 return;
409 }
410
411 if (!failed) {
412 /* good: clear consecutive_errors */
413 s->consecutive_errors = 0;
414 return;
415 }
416
Christopher Faulet29f77e82017-06-08 14:04:45 +0200417 HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100418
419 if (s->consecutive_errors < s->consecutive_errors_limit)
420 return;
421
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100422 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
423 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100424
425 switch (s->onerror) {
426 case HANA_ONERR_FASTINTER:
427 /* force fastinter - nothing to do here as all modes force it */
428 break;
429
430 case HANA_ONERR_SUDDTH:
431 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900432 if (s->check.health > s->check.rise)
433 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100434
435 /* no break - fall through */
436
437 case HANA_ONERR_FAILCHK:
438 /* simulate a failed health check */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200439 set_server_check_status(&s->check, HCHK_STATUS_HANA,
440 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200441 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100442 break;
443
444 case HANA_ONERR_MARKDWN:
445 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900446 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200447 set_server_check_status(&s->check, HCHK_STATUS_HANA,
448 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200449 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100450 break;
451
452 default:
453 /* write a warning? */
454 break;
455 }
456
457 s->consecutive_errors = 0;
Christopher Faulet29f77e82017-06-08 14:04:45 +0200458 HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100459
Simon Horman66183002013-02-23 10:16:43 +0900460 if (s->check.fastinter) {
461 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300462 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200463 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300464 /* requeue check task with new expire */
465 task_queue(s->check.task);
466 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100467 }
468}
469
Willy Tarreaua1dab552014-04-14 15:04:54 +0200470static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100471{
472 int sv_state;
473 int ratio;
474 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800475 char addr[46];
476 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100477 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
478 "UP %d/%d", "UP",
479 "NOLB %d/%d", "NOLB",
480 "no check" };
481
482 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
483 hlen += 24;
484
Willy Tarreauff5ae352013-12-11 20:36:34 +0100485 if (!(s->check.state & CHK_ST_ENABLED))
486 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200487 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900488 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100489 sv_state = 3; /* UP */
490 else
491 sv_state = 2; /* going down */
492
Emeric Brun52a91d32017-08-31 14:41:55 +0200493 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100494 sv_state += 2;
495 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900496 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100497 sv_state = 1; /* going up */
498 else
499 sv_state = 0; /* DOWN */
500 }
501
Willy Tarreaua1dab552014-04-14 15:04:54 +0200502 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100503 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200504 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
505 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100506
Joseph Lynch514061c2015-01-15 17:52:59 -0800507 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100508 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
509 snprintf(port, sizeof(port), "%u", s->svc_port);
510 else
511 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800512
513 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
514 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100515 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200516 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100517 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
518 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
519 s->nbpend);
520
Emeric Brun52a91d32017-08-31 14:41:55 +0200521 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100522 now.tv_sec < s->last_change + s->slowstart &&
523 now.tv_sec >= s->last_change) {
524 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200525 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100526 }
527
528 buffer[hlen++] = '\r';
529 buffer[hlen++] = '\n';
530
531 return hlen;
532}
533
Willy Tarreau20a18342013-12-05 00:31:46 +0100534/* Check the connection. If an error has already been reported or the socket is
535 * closed, keep errno intact as it is supposed to contain the valid error code.
536 * If no error is reported, check the socket's error queue using getsockopt().
537 * Warning, this must be done only once when returning from poll, and never
538 * after an I/O error was attempted, otherwise the error queue might contain
539 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
540 * socket. Returns non-zero if an error was reported, zero if everything is
541 * clean (including a properly closed socket).
542 */
543static int retrieve_errno_from_socket(struct connection *conn)
544{
545 int skerr;
546 socklen_t lskerr = sizeof(skerr);
547
548 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
549 return 1;
550
Willy Tarreau3c728722014-01-23 13:50:42 +0100551 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100552 return 0;
553
Willy Tarreau585744b2017-08-24 14:31:19 +0200554 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100555 errno = skerr;
556
557 if (errno == EAGAIN)
558 errno = 0;
559
560 if (!errno) {
561 /* we could not retrieve an error, that does not mean there is
562 * none. Just don't change anything and only report the prior
563 * error if any.
564 */
565 if (conn->flags & CO_FL_ERROR)
566 return 1;
567 else
568 return 0;
569 }
570
571 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
572 return 1;
573}
574
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100575/* Try to collect as much information as possible on the connection status,
576 * and adjust the server status accordingly. It may make use of <errno_bck>
577 * if non-null when the caller is absolutely certain of its validity (eg:
578 * checked just after a syscall). If the caller doesn't have a valid errno,
579 * it can pass zero, and retrieve_errno_from_socket() will be called to try
580 * to extract errno from the socket. If no error is reported, it will consider
581 * the <expired> flag. This is intended to be used when a connection error was
582 * reported in conn->flags or when a timeout was reported in <expired>. The
583 * function takes care of not updating a server status which was already set.
584 * All situations where at least one of <expired> or CO_FL_ERROR are set
585 * produce a status.
586 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200587static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100588{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200589 struct conn_stream *cs = check->cs;
590 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100591 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200592 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200593 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200594 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100595
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100596 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100597 return;
598
599 errno = errno_bck;
Willy Tarreau00149122017-10-04 18:05:01 +0200600 if (conn && (!errno || errno == EAGAIN))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100601 retrieve_errno_from_socket(conn);
602
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200603 if (conn && !(conn->flags & CO_FL_ERROR) &&
604 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100605 return;
606
607 /* we'll try to build a meaningful error message depending on the
608 * context of the error possibly present in conn->err_code, and the
609 * socket error possibly collected above. This is useful to know the
610 * exact step of the L6 layer (eg: SSL handshake).
611 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200612 chk = get_trash_chunk();
613
614 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900615 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200616 if (!step)
617 chunk_printf(chk, " at initial connection step of tcp-check");
618 else {
619 chunk_printf(chk, " at step %d of tcp-check", step);
620 /* we were looking for a string */
621 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
622 if (check->last_started_step->port)
623 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
624 else
625 chunk_appendf(chk, " (connect)");
626 }
627 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
628 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200629 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200630 else if (check->last_started_step->expect_regex)
631 chunk_appendf(chk, " (expect regex)");
632 }
633 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
634 chunk_appendf(chk, " (send)");
635 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200636
637 comment = tcpcheck_get_step_comment(check, step);
638 if (comment)
639 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200640 }
641 }
642
Willy Tarreau00149122017-10-04 18:05:01 +0200643 if (conn && conn->err_code) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100644 if (errno && errno != EAGAIN)
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200645 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
646 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100647 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200648 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
649 chk->area);
650 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100651 }
652 else {
653 if (errno && errno != EAGAIN) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200654 chunk_printf(&trash, "%s%s", strerror(errno),
655 chk->area);
656 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100657 }
658 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200659 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100660 }
661 }
662
Willy Tarreau00149122017-10-04 18:05:01 +0200663 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200664 /* NOTE: this is reported after <fall> tries */
665 chunk_printf(chk, "No port available for the TCP connection");
666 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
667 }
668
Willy Tarreau00149122017-10-04 18:05:01 +0200669 if (!conn) {
670 /* connection allocation error before the connection was established */
671 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
672 }
673 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100674 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200675 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100676 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
677 else if (expired)
678 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200679
680 /*
681 * might be due to a server IP change.
682 * Let's trigger a DNS resolution if none are currently running.
683 */
Christopher Faulet67957bd2017-09-27 11:00:59 +0200684 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200685
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100686 }
687 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
688 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200689 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100690 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
691 else if (expired)
692 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
693 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200694 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100695 /* I/O error after connection was established and before we could diagnose */
696 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
697 }
698 else if (expired) {
699 /* connection established but expired check */
700 if (check->type == PR_O2_SSL3_CHK)
701 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
702 else /* HTTP, SMTP, ... */
703 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
704 }
705
706 return;
707}
708
Olivier Houchard5c110b92018-08-14 17:04:58 +0200709/* This function checks if any I/O is wanted, and if so, attempts to do so */
710static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200711{
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200712 struct check *check = ctx;
713 struct conn_stream *cs = check->cs;
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200714
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100715 if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200716 wake_srv_chk(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100717 if (!(check->wait_list.events & SUB_RETRY_RECV)) {
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200718 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
719 __event_srv_chk_r(cs);
720 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
721 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200722 return NULL;
723}
724
725/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100726 *
727 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200728 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
729 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200730 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200731static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200732{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200733 struct connection *conn = cs->conn;
734 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900735 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900736 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200737
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100738 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100739 goto out_wakeup;
740
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200741 if (conn->flags & CO_FL_HANDSHAKE) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100742 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200743 goto out;
744 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100745
Willy Tarreau20a18342013-12-05 00:31:46 +0100746 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200747 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100748 goto out_wakeup;
749 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100750
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200751 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100752 /* if the output is closed, we can't do anything */
753 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200754 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100755 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200756 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200757
Willy Tarreau06559ac2013-12-05 01:53:08 +0100758 /* here, we know that the connection is established. That's enough for
759 * a pure TCP check.
760 */
761 if (!check->type)
762 goto out_wakeup;
763
Willy Tarreauc09572f2017-10-04 11:58:22 +0200764 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100765 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200766 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200767
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200768 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200769 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200770 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200771 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200772 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100773 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200774 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200775 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100776 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200777 goto out;
778 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100779 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200780
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100781 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
782 if (s->proxy->timeout.check) {
783 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
784 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200785 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200786 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100787
Willy Tarreau83749182007-04-15 20:56:27 +0200788 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200789 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200790 out:
791 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200792}
793
Willy Tarreaubaaee002006-06-26 02:48:02 +0200794/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200795 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200796 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900797 * set_server_check_status() to update check->status, check->duration
798 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200799
800 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
801 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
802 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
803 * response to an SSL HELLO (the principle is that this is enough to
804 * distinguish between an SSL server and a pure TCP relay). All other cases will
805 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
806 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100807 *
808 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200809 * via the out label.
810 *
811 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200812 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200813static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200814{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200815 struct connection *conn = cs->conn;
816 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900817 struct server *s = check->server;
818 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200819 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100820 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200821 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200822
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100823 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200824 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200825
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200826 if (conn->flags & CO_FL_HANDSHAKE) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100827 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200828 goto out;
829 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200830
Willy Tarreauc09572f2017-10-04 11:58:22 +0200831 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100832 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200833 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200834
Willy Tarreau83749182007-04-15 20:56:27 +0200835 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
836 * but the connection was closed on the remote end. Fortunately, recv still
837 * works correctly and we don't need to do the getsockopt() on linux.
838 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000839
840 /* Set buffer to point to the end of the data already read, and check
841 * that there is free space remaining. If the buffer is full, proceed
842 * with running the checks without attempting another socket read.
843 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000844
Willy Tarreau03938182010-03-17 21:52:07 +0100845 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000846
Olivier Houchard511efea2018-08-16 15:30:32 +0200847 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200848 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100849 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200850 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200851 /* Report network errors only if we got no other data. Otherwise
852 * we'll let the upper layers decide whether the response is OK
853 * or not. It is very common that an RST sent by the server is
854 * reported as an error just after the last data chunk.
855 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200856 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100857 goto out_wakeup;
858 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200859 }
860
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100861
Willy Tarreau03938182010-03-17 21:52:07 +0100862 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200863 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100864 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200865 if (b_data(&check->bi) < b_size(&check->bi))
866 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100867 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200868 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100869 done = 1; /* buffer full, don't wait for more data */
870 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200871
Nick Chalk57b1bf72010-03-16 15:50:46 +0000872 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900873 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200874 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200875 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100876 goto wait_more_data;
877
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100878 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200879 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
880 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
881 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
882 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
883 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
884 cut_crlf(b_head(&check->bi));
885 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200886
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100887 goto out_wakeup;
888 }
889
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200890 check->code = str2uic(b_head(&check->bi) + 9);
891 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200892
Willy Tarreaubd741542010-03-16 18:46:54 +0100893 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200894 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000895 /* 404 may be accepted as "stopping" only if the server was up */
896 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900897 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000898 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100899 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
900 /* Run content verification check... We know we have at least 13 chars */
901 if (!httpchk_expect(s, done))
902 goto wait_more_data;
903 }
904 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200905 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100906 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900907 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100908 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000909 else {
910 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900911 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000912 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200913 break;
914
915 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200916 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100917 goto wait_more_data;
918
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100919 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200920 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900921 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200922 else
Simon Horman4a741432013-02-23 15:35:38 +0900923 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200924 break;
925
926 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200927 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100928 goto wait_more_data;
929
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200930 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200931 if ((b_data(&check->bi) < strlen("000\r")) ||
932 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
933 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
934 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
935 cut_crlf(b_head(&check->bi));
936 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200937 goto out_wakeup;
938 }
939
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200940 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200941
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200942 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200943 cut_crlf(desc);
944
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100945 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200946 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900947 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200948 else
Simon Horman4a741432013-02-23 15:35:38 +0900949 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200950 break;
951
Simon Hormana2b9dad2013-02-12 10:45:54 +0900952 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100953 int status = HCHK_STATUS_CHECKED;
954 const char *hs = NULL; /* health status */
955 const char *as = NULL; /* admin status */
956 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200957 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100958 const char *err = NULL; /* first error to report */
959 const char *wrn = NULL; /* first warning to report */
960 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900961
Willy Tarreau81f5d942013-12-09 20:51:51 +0100962 /* We're getting an agent check response. The agent could
963 * have been disabled in the mean time with a long check
964 * still pending. It is important that we ignore the whole
965 * response.
966 */
967 if (!(check->server->agent.state & CHK_ST_ENABLED))
968 break;
969
970 /* The agent supports strings made of a single line ended by the
971 * first CR ('\r') or LF ('\n'). This line is composed of words
972 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
973 * line may optionally contained a description of a state change
974 * after a sharp ('#'), which is only considered if a health state
975 * is announced.
976 *
977 * Words may be composed of :
978 * - a numeric weight suffixed by the percent character ('%').
979 * - a health status among "up", "down", "stopped", and "fail".
980 * - an admin status among "ready", "drain", "maint".
981 *
982 * These words may appear in any order. If multiple words of the
983 * same category appear, the last one wins.
984 */
985
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200986 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +0100987 while (*p && *p != '\n' && *p != '\r')
988 p++;
989
990 if (!*p) {
991 if (!done)
992 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900993
Willy Tarreau9809b782013-12-11 21:40:11 +0100994 /* at least inform the admin that the agent is mis-behaving */
995 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
996 break;
997 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100998
Willy Tarreau9809b782013-12-11 21:40:11 +0100999 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001000 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001001
Willy Tarreau81f5d942013-12-09 20:51:51 +01001002 while (*cmd) {
1003 /* look for next word */
1004 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1005 cmd++;
1006 continue;
1007 }
Simon Horman671b6f02013-11-25 10:46:39 +09001008
Willy Tarreau81f5d942013-12-09 20:51:51 +01001009 if (*cmd == '#') {
1010 /* this is the beginning of a health status description,
1011 * skip the sharp and blanks.
1012 */
1013 cmd++;
1014 while (*cmd == '\t' || *cmd == ' ')
1015 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001016 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001017 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001018
1019 /* find the end of the word so that we have a null-terminated
1020 * word between <cmd> and <p>.
1021 */
1022 p = cmd + 1;
1023 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1024 p++;
1025 if (*p)
1026 *p++ = 0;
1027
1028 /* first, health statuses */
1029 if (strcasecmp(cmd, "up") == 0) {
1030 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001031 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001032 hs = cmd;
1033 }
1034 else if (strcasecmp(cmd, "down") == 0) {
1035 check->health = 0;
1036 status = HCHK_STATUS_L7STS;
1037 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001038 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001039 else if (strcasecmp(cmd, "stopped") == 0) {
1040 check->health = 0;
1041 status = HCHK_STATUS_L7STS;
1042 hs = cmd;
1043 }
1044 else if (strcasecmp(cmd, "fail") == 0) {
1045 check->health = 0;
1046 status = HCHK_STATUS_L7STS;
1047 hs = cmd;
1048 }
1049 /* admin statuses */
1050 else if (strcasecmp(cmd, "ready") == 0) {
1051 as = cmd;
1052 }
1053 else if (strcasecmp(cmd, "drain") == 0) {
1054 as = cmd;
1055 }
1056 else if (strcasecmp(cmd, "maint") == 0) {
1057 as = cmd;
1058 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001059 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001060 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1061 ps = cmd;
1062 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001063 /* try to parse a maxconn here */
1064 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1065 cs = cmd;
1066 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001067 else {
1068 /* keep a copy of the first error */
1069 if (!err)
1070 err = cmd;
1071 }
1072 /* skip to next word */
1073 cmd = p;
1074 }
1075 /* here, cmd points either to \0 or to the beginning of a
1076 * description. Skip possible leading spaces.
1077 */
1078 while (*cmd == ' ' || *cmd == '\n')
1079 cmd++;
1080
1081 /* First, update the admin status so that we avoid sending other
1082 * possibly useless warnings and can also update the health if
1083 * present after going back up.
1084 */
1085 if (as) {
1086 if (strcasecmp(as, "drain") == 0)
1087 srv_adm_set_drain(check->server);
1088 else if (strcasecmp(as, "maint") == 0)
1089 srv_adm_set_maint(check->server);
1090 else
1091 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001092 }
1093
Willy Tarreau81f5d942013-12-09 20:51:51 +01001094 /* now change weights */
1095 if (ps) {
1096 const char *msg;
1097
1098 msg = server_parse_weight_change_request(s, ps);
1099 if (!wrn || !*wrn)
1100 wrn = msg;
1101 }
1102
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001103 if (cs) {
1104 const char *msg;
1105
1106 cs += strlen("maxconn:");
1107
1108 msg = server_parse_maxconn_change_request(s, cs);
1109 if (!wrn || !*wrn)
1110 wrn = msg;
1111 }
1112
Willy Tarreau81f5d942013-12-09 20:51:51 +01001113 /* and finally health status */
1114 if (hs) {
1115 /* We'll report some of the warnings and errors we have
1116 * here. Down reports are critical, we leave them untouched.
1117 * Lack of report, or report of 'UP' leaves the room for
1118 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001119 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001120 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001121 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001122
1123 if (!*msg || status == HCHK_STATUS_L7OKD) {
1124 if (err && *err)
1125 msg = err;
1126 else if (wrn && *wrn)
1127 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001128 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001129
1130 t = get_trash_chunk();
1131 chunk_printf(t, "via agent : %s%s%s%s",
1132 hs, *msg ? " (" : "",
1133 msg, *msg ? ")" : "");
1134
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001135 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001136 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001137 else if (err && *err) {
1138 /* No status change but we'd like to report something odd.
1139 * Just report the current state and copy the message.
1140 */
1141 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001142 set_server_check_status(check, status/*check->status*/,
1143 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001144
Willy Tarreau81f5d942013-12-09 20:51:51 +01001145 }
1146 else if (wrn && *wrn) {
1147 /* No status change but we'd like to report something odd.
1148 * Just report the current state and copy the message.
1149 */
1150 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001151 set_server_check_status(check, status/*check->status*/,
1152 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001153 }
1154 else
1155 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001156 break;
1157 }
1158
Willy Tarreau1620ec32011-08-06 17:05:02 +02001159 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001160 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001161 goto wait_more_data;
1162
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001163 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001164 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001165 }
1166 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001167 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1168 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001169 else
1170 desc = "PostgreSQL unknown error";
1171
Simon Horman4a741432013-02-23 15:35:38 +09001172 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001173 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001174 break;
1175
1176 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001177 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001178 goto wait_more_data;
1179
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001180 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001181 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001182 }
1183 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001184 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001185 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001186 break;
1187
1188 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001189 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001190 goto wait_more_data;
1191
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001192 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001193 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001194 /* We set the MySQL Version in description for information purpose
1195 * FIXME : it can be cool to use MySQL Version for other purpose,
1196 * like mark as down old MySQL server.
1197 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001198 if (b_data(&check->bi) > 51) {
1199 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001200 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001201 }
1202 else {
1203 if (!done)
1204 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001205
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001206 /* it seems we have a OK packet but without a valid length,
1207 * it must be a protocol error
1208 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001209 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001210 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001211 }
1212 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001213 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001214 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001215 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001216 }
1217 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001218 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1219 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1220 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001221
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001222 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001223 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001224 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001225 /* We have only one MySQL packet and it is a Handshake Initialization packet
1226 * but we need to have a second packet to know if it is alright
1227 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001228 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229 goto wait_more_data;
1230 }
1231 else {
1232 /* We have only one packet and it is an Error packet,
1233 * an error message is attached, so we can display it
1234 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001235 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001236 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001237 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001238 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001239 } else if (b_data(&check->bi) > first_packet_len + 4) {
1240 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1241 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1242 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001243
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001244 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001245 /* We have 2 packets and that's good */
1246 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001247 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001248 /* No error packet */
1249 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001250 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001251 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001252 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001253 }
1254 else {
1255 /* An error message is attached in the Error packet
1256 * so we can display it ! :)
1257 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001258 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001259 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001260 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001261 }
1262 }
1263 }
1264 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001265 if (!done)
1266 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001267
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001268 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001269 * it must be a protocol error
1270 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001271 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001272 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001273 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001274 }
1275 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001276 break;
1277
1278 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001279 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001280 goto wait_more_data;
1281
1282 /* Check if the server speaks LDAP (ASN.1/BER)
1283 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1284 * http://tools.ietf.org/html/rfc4511
1285 */
1286
1287 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1288 * LDAPMessage: 0x30: SEQUENCE
1289 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001290 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001291 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001292 }
1293 else {
1294 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001295 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001296
1297 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1298 * messageID: 0x02 0x01 0x01: INTEGER 1
1299 * protocolOp: 0x61: bindResponse
1300 */
1301 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001302 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001303 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001304 goto out_wakeup;
1305 }
1306
1307 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001308 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001309
1310 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1311 * ldapResult: 0x0a 0x01: ENUMERATION
1312 */
1313 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001314 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001315 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001316 goto out_wakeup;
1317 }
1318
1319 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1320 * resultCode
1321 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001322 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001323 if (check->code) {
1324 set_server_check_status(check, HCHK_STATUS_L7STS, "See RFC: http://tools.ietf.org/html/rfc4511#section-4.1.9");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001325 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001326 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001327 }
1328 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001329 break;
1330
Christopher Fauletba7bc162016-11-07 21:07:38 +01001331 case PR_O2_SPOP_CHK: {
1332 unsigned int framesz;
1333 char err[HCHK_DESC_LEN];
1334
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001335 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001336 goto wait_more_data;
1337
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001338 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001339 framesz = ntohl(framesz);
1340
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001341 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001342 goto wait_more_data;
1343
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001344 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001345 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1346 else
1347 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1348 break;
1349 }
1350
Willy Tarreau1620ec32011-08-06 17:05:02 +02001351 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001352 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001353 break;
1354 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001355
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001356 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001357 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001358 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001359 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001360
Nick Chalk57b1bf72010-03-16 15:50:46 +00001361 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001362 *b_head(&check->bi) = '\0';
1363 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001364
Steven Davidovitz544d4812017-03-08 11:06:20 -08001365 /* Close the connection... We still attempt to nicely close if,
1366 * for instance, SSL needs to send a "close notify." Later, we perform
1367 * a hard close and reset the connection if some data are pending,
1368 * otherwise we end up with many TIME_WAITs and eat all the source port
1369 * range quickly. To avoid sending RSTs all the time, we first try to
1370 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001371 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001372 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001373
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001374 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001375 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001376 conn->flags |= CO_FL_ERROR;
1377
Willy Tarreaufdccded2008-08-29 18:19:04 +02001378 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001379out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001380 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001381
1382 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001383 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001384 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001385}
1386
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001387/*
1388 * This function is used only for server health-checks. It handles connection
1389 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001390 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1391 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001392 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001393static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001394{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001395 struct connection *conn = cs->conn;
1396 struct check *check = cs->data;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001397 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001398
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001399 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001400
Willy Tarreauc09572f2017-10-04 11:58:22 +02001401 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001402 if (check->type == PR_O2_TCPCHK_CHK) {
1403 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001404 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001405 conn = cs->conn;
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001406 } else if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001407 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001408
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001409 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001410 /* We may get error reports bypassing the I/O handlers, typically
1411 * the case when sending a pure TCP check which fails, then the I/O
1412 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001413 * main processing task so let's simply wake it up. If we get here,
1414 * we expect errno to still be valid.
1415 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001416 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001417 task_wakeup(check->task, TASK_WOKEN_IO);
1418 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001419 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001420 /* we may get here if only a connection probe was required : we
1421 * don't have any data to send nor anything expected in response,
1422 * so the completion of the connection establishment is enough.
1423 */
1424 task_wakeup(check->task, TASK_WOKEN_IO);
1425 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001426
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001427 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001428 /* We're here because nobody wants to handle the error, so we
1429 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001430 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001431 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001432 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001433 ret = -1;
Willy Tarreau2d351b62013-12-05 02:36:25 +01001434 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001435
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001436 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001437
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001438 /* if a connection got replaced, we must absolutely prevent the connection
1439 * handler from touching its fd, and perform the FD polling updates ourselves
1440 */
1441 if (ret < 0)
1442 conn_cond_update_polling(conn);
1443
1444 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001445}
1446
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001447struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001448 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001449 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001450};
1451
Willy Tarreaubaaee002006-06-26 02:48:02 +02001452/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001453 * updates the server's weight during a warmup stage. Once the final weight is
1454 * reached, the task automatically stops. Note that any server status change
1455 * must have updated s->last_change accordingly.
1456 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001457static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001458{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001459 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001460
1461 /* by default, plan on stopping the task */
1462 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001463 if ((s->next_admin & SRV_ADMF_MAINT) ||
1464 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001465 return t;
1466
Willy Tarreau892337c2014-05-13 23:41:20 +02001467 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001468 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001469
1470 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001471 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001472
1473 /* get back there in 1 second or 1/20th of the slowstart interval,
1474 * whichever is greater, resulting in small 5% steps.
1475 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001476 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001477 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1478 return t;
1479}
1480
Willy Tarreau894c6422017-10-04 15:58:52 +02001481/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1482 * none was found.
1483 */
1484static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1485{
1486 struct tcpcheck_rule *r;
1487
1488 list_for_each_entry(r, list, list) {
1489 if (r->action != TCPCHK_ACT_COMMENT)
1490 return r;
1491 }
1492 return NULL;
1493}
1494
Willy Tarreau2e993902011-10-31 11:53:20 +01001495/*
Simon Horman98637e52014-06-20 12:30:16 +09001496 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001497 *
1498 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001499 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1500 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1501 * - SF_ERR_SRVTO if there are no more servers
1502 * - SF_ERR_SRVCL if the connection was refused by the server
1503 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1504 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1505 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001506 * - SF_ERR_CHK_PORT if no port could be found to run a health check on an AF_INET* socket
Tim Düsterhus4896c442016-11-29 02:15:19 +01001507 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001508 * Note that we try to prevent the network stack from sending the ACK during the
1509 * connect() when a pure TCP check is used (without PROXY protocol).
1510 */
Simon Horman98637e52014-06-20 12:30:16 +09001511static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001512{
1513 struct check *check = t->context;
1514 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001515 struct conn_stream *cs = check->cs;
1516 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001517 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001518 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001519 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001520 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001521
Willy Tarreau00149122017-10-04 18:05:01 +02001522 /* we cannot have a connection here */
1523 if (conn)
1524 return SF_ERR_INTERNAL;
1525
Simon Hormanb00d17a2014-06-13 16:18:16 +09001526 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001527 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001528 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001529 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1530 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001531
1532 /* prepare the check buffer.
1533 * This should not be used if check is the secondary agent check
1534 * of a server as s->proxy->check_req will relate to the
1535 * configuration of the primary check. Similarly, tcp-check uses
1536 * its own strings.
1537 */
1538 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001539 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001540
1541 /* we want to check if this host replies to HTTP or SSLv3 requests
1542 * so we'll send the request, and won't wake the checker up now.
1543 */
1544 if ((check->type) == PR_O2_SSL3_CHK) {
1545 /* SSL requires that we put Unix time in the request */
1546 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001547 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001548 }
1549 else if ((check->type) == PR_O2_HTTP_CHK) {
1550 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001551 b_putblk(&check->bo, trash.area,
1552 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001553 /* prevent HTTP keep-alive when "http-check expect" is used */
1554 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001555 b_putist(&check->bo, ist("Connection: close\r\n"));
1556 b_putist(&check->bo, ist("\r\n"));
1557 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001558 }
1559 }
1560
James Brown55f9ff12015-10-21 18:19:05 -07001561 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001562 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001563 }
1564
Willy Tarreauf411cce2017-10-04 16:21:19 +02001565 /* for tcp-checks, the initial connection setup is handled separately as
1566 * it may be sent to a specific port and not to the server's.
1567 */
1568 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1569 tcpcheck_main(check);
1570 return SF_ERR_UP;
1571 }
1572
Simon Hormanb00d17a2014-06-13 16:18:16 +09001573 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001574 cs = check->cs = cs_new(NULL);
1575 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001576 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001577 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001578 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001579 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001580
Simon Horman41f58762015-01-30 11:22:56 +09001581 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001582 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001583 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001584 }
1585 else {
1586 /* we'll connect to the addr on the server */
1587 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001588 }
1589
Olivier Houchard6377a002017-12-01 22:04:05 +01001590 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001591 conn->target = &s->obj_type;
1592
Willy Tarreaube373152018-09-06 11:45:30 +02001593 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001594 int i = 0;
1595
1596 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001597 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001598 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001599
1600 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001601 }
1602
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001603 /* no client address */
1604 clear_addr(&conn->addr.from);
1605
Willy Tarreaube373152018-09-06 11:45:30 +02001606 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf502aca2018-12-14 19:42:40 +01001607 conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL);
Willy Tarreaube373152018-09-06 11:45:30 +02001608 cs_attach(cs, check, &check_conn_cb);
1609
Willy Tarreauf3d34822014-12-08 12:11:28 +01001610 /* only plain tcp-check supports quick ACK */
1611 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1612
Willy Tarreauf411cce2017-10-04 16:21:19 +02001613 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT)
1614 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001615
Willy Tarreaue7dff022015-04-03 01:14:29 +02001616 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001617 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001618 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Willy Tarreau16257f62017-11-02 15:45:00 +01001619
Willy Tarreau16257f62017-11-02 15:45:00 +01001620
Olivier Houchard9130a962017-10-17 17:33:43 +02001621#ifdef USE_OPENSSL
1622 if (s->check.sni)
1623 ssl_sock_set_servername(conn, s->check.sni);
1624#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001625 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001626 conn->send_proxy_ofs = 1;
1627 conn->flags |= CO_FL_SEND_PROXY;
1628 }
1629
1630 return ret;
1631}
1632
Simon Horman98637e52014-06-20 12:30:16 +09001633static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001634static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001635__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001636
1637void block_sigchld(void)
1638{
1639 sigset_t set;
1640 sigemptyset(&set);
1641 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001642 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001643}
1644
1645void unblock_sigchld(void)
1646{
1647 sigset_t set;
1648 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001649 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001650 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001651}
1652
Simon Horman98637e52014-06-20 12:30:16 +09001653static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1654{
1655 struct pid_list *elem;
1656 struct check *check = t->context;
1657
Willy Tarreaubafbe012017-11-24 17:34:44 +01001658 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001659 if (!elem)
1660 return NULL;
1661 elem->pid = pid;
1662 elem->t = t;
1663 elem->exited = 0;
1664 check->curpid = elem;
1665 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001666
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001667 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001668 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001669 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001670
Simon Horman98637e52014-06-20 12:30:16 +09001671 return elem;
1672}
1673
Simon Horman98637e52014-06-20 12:30:16 +09001674static void pid_list_del(struct pid_list *elem)
1675{
1676 struct check *check;
1677
1678 if (!elem)
1679 return;
1680
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001681 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001682 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001683 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001684
Simon Horman98637e52014-06-20 12:30:16 +09001685 if (!elem->exited)
1686 kill(elem->pid, SIGTERM);
1687
1688 check = elem->t->context;
1689 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001690 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001691}
1692
1693/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1694static void pid_list_expire(pid_t pid, int status)
1695{
1696 struct pid_list *elem;
1697
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001698 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001699 list_for_each_entry(elem, &pid_list, list) {
1700 if (elem->pid == pid) {
1701 elem->t->expire = now_ms;
1702 elem->status = status;
1703 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001704 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001705 break;
Simon Horman98637e52014-06-20 12:30:16 +09001706 }
1707 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001708 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001709}
1710
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001711static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001712{
1713 pid_t pid;
1714 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001715
Simon Horman98637e52014-06-20 12:30:16 +09001716 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1717 pid_list_expire(pid, status);
1718}
1719
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001720static int init_pid_list(void)
1721{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001722 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001723 /* Nothing to do */
1724 return 0;
1725
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001726 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001727 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1728 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001729 return 1;
1730 }
1731
Willy Tarreaubafbe012017-11-24 17:34:44 +01001732 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1733 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001734 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1735 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001736 return 1;
1737 }
1738
1739 return 0;
1740}
1741
Cyril Bontéac92a062014-12-27 22:28:38 +01001742/* helper macro to set an environment variable and jump to a specific label on failure. */
1743#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001744
1745/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001746 * helper function to allocate enough memory to store an environment variable.
1747 * It will also check that the environment variable is updatable, and silently
1748 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001749 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001750static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001751{
1752 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001753 char *envname;
1754 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001755
Cyril Bontéac92a062014-12-27 22:28:38 +01001756 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001757 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001758 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001759 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001760
1761 envname = extcheck_envs[idx].name;
1762 vmaxlen = extcheck_envs[idx].vmaxlen;
1763
1764 /* Check if the environment variable is already set, and silently reject
1765 * the update if this one is not updatable. */
1766 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1767 return 0;
1768
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001769 /* Instead of sending NOT_USED, sending an empty value is preferable */
1770 if (strcmp(value, "NOT_USED") == 0) {
1771 value = "";
1772 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001773
1774 len = strlen(envname) + 1;
1775 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1776 len += strlen(value);
1777 else
1778 len += vmaxlen;
1779
1780 if (!check->envp[idx])
1781 check->envp[idx] = malloc(len + 1);
1782
1783 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001784 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001785 return 1;
1786 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001787 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001788 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001789 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001790 return 1;
1791 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001792 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001793 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001794 return 1;
1795 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001796 return 0;
1797}
Simon Horman98637e52014-06-20 12:30:16 +09001798
1799static int prepare_external_check(struct check *check)
1800{
1801 struct server *s = check->server;
1802 struct proxy *px = s->proxy;
1803 struct listener *listener = NULL, *l;
1804 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001805 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001806 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001807
1808 list_for_each_entry(l, &px->conf.listeners, by_fe)
1809 /* Use the first INET, INET6 or UNIX listener */
1810 if (l->addr.ss_family == AF_INET ||
1811 l->addr.ss_family == AF_INET6 ||
1812 l->addr.ss_family == AF_UNIX) {
1813 listener = l;
1814 break;
1815 }
1816
Simon Horman98637e52014-06-20 12:30:16 +09001817 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001818 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1819 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001820 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001821 goto err;
1822 }
Simon Horman98637e52014-06-20 12:30:16 +09001823
Cyril Bontéac92a062014-12-27 22:28:38 +01001824 check->argv = calloc(6, sizeof(char *));
1825 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001826 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001827 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001828 }
Simon Horman98637e52014-06-20 12:30:16 +09001829
1830 check->argv[0] = px->check_command;
1831
Cyril Bonté777be862014-12-02 21:21:35 +01001832 if (!listener) {
1833 check->argv[1] = strdup("NOT_USED");
1834 check->argv[2] = strdup("NOT_USED");
1835 }
1836 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001837 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001838 addr_to_str(&listener->addr, buf, sizeof(buf));
1839 check->argv[1] = strdup(buf);
1840 port_to_str(&listener->addr, buf, sizeof(buf));
1841 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001842 }
1843 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001844 const struct sockaddr_un *un;
1845
1846 un = (struct sockaddr_un *)&listener->addr;
1847 check->argv[1] = strdup(un->sun_path);
1848 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001849 }
1850 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001851 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001852 goto err;
1853 }
1854
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001855 addr_to_str(&s->addr, buf, sizeof(buf));
1856 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001857
1858 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1859 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1860 else
1861 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001862 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001863
Cyril Bontéac92a062014-12-27 22:28:38 +01001864 for (i = 0; i < 5; i++) {
1865 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001866 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001867 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001868 }
1869 }
Simon Horman98637e52014-06-20 12:30:16 +09001870
Cyril Bontéac92a062014-12-27 22:28:38 +01001871 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001872 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001873 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1874 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1875 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1876 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001877 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001878 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1879 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1880 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1881 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1882 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1883 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1884
1885 /* Ensure that we don't leave any hole in check->envp */
1886 for (i = 0; i < EXTCHK_SIZE; i++)
1887 if (!check->envp[i])
1888 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001889
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001890 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001891err:
1892 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001893 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001894 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001895 free(check->envp);
1896 check->envp = NULL;
1897 }
1898
1899 if (check->argv) {
1900 for (i = 1; i < 5; i++)
1901 free(check->argv[i]);
1902 free(check->argv);
1903 check->argv = NULL;
1904 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001905 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001906}
1907
Simon Hormanb00d17a2014-06-13 16:18:16 +09001908/*
Simon Horman98637e52014-06-20 12:30:16 +09001909 * establish a server health-check that makes use of a process.
1910 *
1911 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001912 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001913 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001914 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001915 *
1916 * Blocks and then unblocks SIGCHLD
1917 */
1918static int connect_proc_chk(struct task *t)
1919{
Cyril Bontéac92a062014-12-27 22:28:38 +01001920 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001921 struct check *check = t->context;
1922 struct server *s = check->server;
1923 struct proxy *px = s->proxy;
1924 int status;
1925 pid_t pid;
1926
Willy Tarreaue7dff022015-04-03 01:14:29 +02001927 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001928
1929 block_sigchld();
1930
1931 pid = fork();
1932 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001933 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1934 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001935 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1936 goto out;
1937 }
1938 if (pid == 0) {
1939 /* Child */
1940 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001941 int fd;
1942
1943 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1944 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1945
1946 while (fd < global.rlimit_nofile)
1947 close(fd++);
1948
Simon Horman98637e52014-06-20 12:30:16 +09001949 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001950 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001951 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001952 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1953 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001954 exit(-1);
1955 }
1956
1957 /* Parent */
1958 if (check->result == CHK_RES_UNKNOWN) {
1959 if (pid_list_add(pid, t) != NULL) {
1960 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1961
1962 if (px->timeout.check && px->timeout.connect) {
1963 int t_con = tick_add(now_ms, px->timeout.connect);
1964 t->expire = tick_first(t->expire, t_con);
1965 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001966 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001967 goto out;
1968 }
1969 else {
1970 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1971 }
1972 kill(pid, SIGTERM); /* process creation error */
1973 }
1974 else
1975 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1976
1977out:
1978 unblock_sigchld();
1979 return status;
1980}
1981
1982/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02001983 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02001984 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001985 *
1986 * Please do NOT place any return statement in this function and only leave
1987 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001988 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001989static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09001990{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001991 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09001992 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09001993 int rv;
1994 int ret;
1995 int expired = tick_is_expired(t->expire, now_ms);
1996
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001997 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09001998 if (!(check->state & CHK_ST_INPROGRESS)) {
1999 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002000 if (!expired) /* woke up too early */
2001 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002002
2003 /* we don't send any health-checks when the proxy is
2004 * stopped, the server should not be checked or the check
2005 * is disabled.
2006 */
2007 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2008 s->proxy->state == PR_STSTOPPED)
2009 goto reschedule;
2010
2011 /* we'll initiate a new check */
2012 set_server_check_status(check, HCHK_STATUS_START, NULL);
2013
2014 check->state |= CHK_ST_INPROGRESS;
2015
Simon Hormandbf70192015-01-30 11:22:53 +09002016 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002017 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002018 /* the process was forked, we allow up to min(inter,
2019 * timeout.connect) for it to report its status, but
2020 * only when timeout.check is set as it may be to short
2021 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002022 */
2023 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2024
2025 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2026 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2027 t->expire = tick_first(t->expire, t_con);
2028 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002029 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002030 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002031 }
2032
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002033 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002034
2035 check->state &= ~CHK_ST_INPROGRESS;
2036 check_notify_failure(check);
2037
2038 /* we allow up to min(inter, timeout.connect) for a connection
2039 * to establish but only when timeout.check is set
2040 * as it may be to short for a full check otherwise
2041 */
2042 while (tick_is_expired(t->expire, now_ms)) {
2043 int t_con;
2044
2045 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2046 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2047
2048 if (s->proxy->timeout.check)
2049 t->expire = tick_first(t->expire, t_con);
2050 }
2051 }
2052 else {
2053 /* there was a test running.
2054 * First, let's check whether there was an uncaught error,
2055 * which can happen on connect timeout or error.
2056 */
2057 if (check->result == CHK_RES_UNKNOWN) {
2058 /* good connection is enough for pure TCP check */
2059 struct pid_list *elem = check->curpid;
2060 int status = HCHK_STATUS_UNKNOWN;
2061
2062 if (elem->exited) {
2063 status = elem->status; /* Save in case the process exits between use below */
2064 if (!WIFEXITED(status))
2065 check->code = -1;
2066 else
2067 check->code = WEXITSTATUS(status);
2068 if (!WIFEXITED(status) || WEXITSTATUS(status))
2069 status = HCHK_STATUS_PROCERR;
2070 else
2071 status = HCHK_STATUS_PROCOK;
2072 } else if (expired) {
2073 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002074 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002075 kill(elem->pid, SIGTERM);
2076 }
2077 set_server_check_status(check, status, NULL);
2078 }
2079
2080 if (check->result == CHK_RES_FAILED) {
2081 /* a failure or timeout detected */
2082 check_notify_failure(check);
2083 }
2084 else if (check->result == CHK_RES_CONDPASS) {
2085 /* check is OK but asks for stopping mode */
2086 check_notify_stopping(check);
2087 }
2088 else if (check->result == CHK_RES_PASSED) {
2089 /* a success was detected */
2090 check_notify_success(check);
2091 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002092 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002093 check->state &= ~CHK_ST_INPROGRESS;
2094
2095 pid_list_del(check->curpid);
2096
2097 rv = 0;
2098 if (global.spread_checks > 0) {
2099 rv = srv_getinter(check) * global.spread_checks / 100;
2100 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2101 }
2102 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2103 }
2104
2105 reschedule:
2106 while (tick_is_expired(t->expire, now_ms))
2107 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002108
2109 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002110 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002111 return t;
2112}
2113
2114/*
2115 * manages a server health-check that uses a connection. Returns
2116 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002117 *
2118 * Please do NOT place any return statement in this function and only leave
2119 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002120 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002121static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002122{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002123 struct check *check = context;
Simon Horman4a741432013-02-23 15:35:38 +09002124 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002125 struct conn_stream *cs = check->cs;
2126 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002127 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002128 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002129 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002130
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002131 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002132 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002133 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002134 if (!expired) /* woke up too early */
2135 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002136
Simon Horman671b6f02013-11-25 10:46:39 +09002137 /* we don't send any health-checks when the proxy is
2138 * stopped, the server should not be checked or the check
2139 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002140 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002141 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01002142 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002143 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002144
2145 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002146 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002147
Willy Tarreau2c115e52013-12-11 19:41:16 +01002148 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002149 b_reset(&check->bi);
2150 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002151
Simon Hormandbf70192015-01-30 11:22:53 +09002152 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002153 cs = check->cs;
2154 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002155
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002156 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002157 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002158 goto out_unlock;
2159
Willy Tarreaue7dff022015-04-03 01:14:29 +02002160 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002161 /* we allow up to min(inter, timeout.connect) for a connection
2162 * to establish but only when timeout.check is set
2163 * as it may be to short for a full check otherwise
2164 */
Simon Horman4a741432013-02-23 15:35:38 +09002165 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002166
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002167 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2168 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2169 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002170 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002171
Olivier Houchard53216e72018-10-10 15:46:36 +02002172 if (check->type)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002173 __event_srv_chk_r(cs);
Willy Tarreau06559ac2013-12-05 01:53:08 +01002174
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002175 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002176 goto reschedule;
2177
Willy Tarreaue7dff022015-04-03 01:14:29 +02002178 case SF_ERR_SRVTO: /* ETIMEDOUT */
2179 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002180 if (conn)
2181 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002182 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002183 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002184 /* should share same code than cases below */
2185 case SF_ERR_CHK_PORT:
2186 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002187 case SF_ERR_PRXCOND:
2188 case SF_ERR_RESOURCE:
2189 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002190 if (conn)
2191 conn->flags |= CO_FL_ERROR;
2192 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002193 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002194 }
2195
Willy Tarreau5a78f362012-11-23 12:47:05 +01002196 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002197 if (cs) {
2198 cs_destroy(cs);
2199 cs = check->cs = NULL;
2200 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002201 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002202
Willy Tarreau2c115e52013-12-11 19:41:16 +01002203 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002204 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002205
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002206 /* we allow up to min(inter, timeout.connect) for a connection
2207 * to establish but only when timeout.check is set
2208 * as it may be to short for a full check otherwise
2209 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002210 while (tick_is_expired(t->expire, now_ms)) {
2211 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002212
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002213 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002214 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002215
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002216 if (s->proxy->timeout.check)
2217 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002218 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002219 }
2220 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002221 /* there was a test running.
2222 * First, let's check whether there was an uncaught error,
2223 * which can happen on connect timeout or error.
2224 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002225 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002226 /* good connection is enough for pure TCP check */
2227 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002228 if (check->use_ssl)
2229 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002230 else
Simon Horman4a741432013-02-23 15:35:38 +09002231 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002232 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002233 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002234 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002235 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002236 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002237 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002238 }
2239
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002240 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002241 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002242 /* The check was aborted and the connection was not yet closed.
2243 * This can happen upon timeout, or when an external event such
2244 * as a failed response coupled with "observe layer7" caused the
2245 * server state to be suddenly changed.
2246 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002247 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002248 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002249 }
2250
Willy Tarreauac59f362017-10-08 11:10:19 +02002251 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002252 cs_destroy(cs);
2253 cs = check->cs = NULL;
2254 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002255 }
2256
Willy Tarreauaf549582014-05-16 17:37:50 +02002257 if (check->result == CHK_RES_FAILED) {
2258 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02002259 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002260 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02002261 else if (check->result == CHK_RES_CONDPASS) {
2262 /* check is OK but asks for stopping mode */
2263 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002264 }
Willy Tarreau3e048382014-05-21 10:30:54 +02002265 else if (check->result == CHK_RES_PASSED) {
2266 /* a success was detected */
2267 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002268 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002269 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002270 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002271
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002272 rv = 0;
2273 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09002274 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002275 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002276 }
Simon Horman4a741432013-02-23 15:35:38 +09002277 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002278 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002279
2280 reschedule:
2281 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002282 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002283 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002284 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002285 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002286}
2287
Simon Horman98637e52014-06-20 12:30:16 +09002288/*
2289 * manages a server health-check. Returns
2290 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2291 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002292static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002293{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002294 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002295
2296 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002297 return process_chk_proc(t, context, state);
2298 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002299
Simon Horman98637e52014-06-20 12:30:16 +09002300}
2301
Simon Horman5c942422013-11-25 10:46:32 +09002302static int start_check_task(struct check *check, int mininter,
2303 int nbcheck, int srvpos)
2304{
2305 struct task *t;
2306 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002307 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002308 ha_alert("Starting [%s:%s] check: out of memory.\n",
2309 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002310 return 0;
2311 }
2312
2313 check->task = t;
2314 t->process = process_chk;
2315 t->context = check;
2316
Willy Tarreau1746eec2014-04-25 10:46:47 +02002317 if (mininter < srv_getinter(check))
2318 mininter = srv_getinter(check);
2319
2320 if (global.max_spread_checks && mininter > global.max_spread_checks)
2321 mininter = global.max_spread_checks;
2322
Simon Horman5c942422013-11-25 10:46:32 +09002323 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002324 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002325 check->start = now;
2326 task_queue(t);
2327
2328 return 1;
2329}
2330
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002331/*
2332 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002333 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002334 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002335static int start_checks()
2336{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002337
2338 struct proxy *px;
2339 struct server *s;
2340 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002341 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002342
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002343 /* 1- count the checkers to run simultaneously.
2344 * We also determine the minimum interval among all of those which
2345 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2346 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002347 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002348 * too short an interval for all others.
2349 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002350 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002351 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002352 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002353 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002354 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002355 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002356 }
2357 /* We need a warmup task that will be called when the server
2358 * state switches from down to up.
2359 */
2360 s->warmup = t;
2361 t->process = server_warmup;
2362 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002363 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002364 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002365 task_schedule(s->warmup, tick_add(now_ms, MS_TO_TICKS(MAX(1000, (now.tv_sec - s->last_change)) / 20)));
Willy Tarreaue7b73482013-11-21 11:50:50 +01002366 }
2367
Willy Tarreaud8514a22013-12-11 21:10:14 +01002368 if (s->check.state & CHK_ST_CONFIGURED) {
2369 nbcheck++;
2370 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2371 (!mininter || mininter > srv_getinter(&s->check)))
2372 mininter = srv_getinter(&s->check);
2373 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002374
Willy Tarreaud8514a22013-12-11 21:10:14 +01002375 if (s->agent.state & CHK_ST_CONFIGURED) {
2376 nbcheck++;
2377 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2378 (!mininter || mininter > srv_getinter(&s->agent)))
2379 mininter = srv_getinter(&s->agent);
2380 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002381 }
2382 }
2383
Simon Horman4a741432013-02-23 15:35:38 +09002384 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002385 return 0;
2386
2387 srand((unsigned)time(NULL));
2388
2389 /*
2390 * 2- start them as far as possible from each others. For this, we will
2391 * start them after their interval set to the min interval divided by
2392 * the number of servers, weighted by the server's position in the list.
2393 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002394 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002395 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2396 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002397 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002398 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002399 }
2400 }
2401
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002402 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002403 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002404 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002405 if (s->check.type == PR_O2_EXT_CHK) {
2406 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002407 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002408 }
Simon Hormand60d6912013-11-25 10:46:36 +09002409 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002410 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002411 srvpos++;
2412 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002413
Simon Hormand60d6912013-11-25 10:46:36 +09002414 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002415 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002416 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002417 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002418 }
2419 srvpos++;
2420 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002421 }
2422 }
2423 return 0;
2424}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002425
2426/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002427 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002428 * The buffer MUST be terminated by a null byte before calling this function.
2429 * Sets server status appropriately. The caller is responsible for ensuring
2430 * that the buffer contains at least 13 characters. If <done> is zero, we may
2431 * return 0 to indicate that data is required to decide of a match.
2432 */
2433static int httpchk_expect(struct server *s, int done)
2434{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002435 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002436 char status_code[] = "000";
2437 char *contentptr;
2438 int crlf;
2439 int ret;
2440
2441 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2442 case PR_O2_EXP_STS:
2443 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002444 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2445 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002446
2447 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2448 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2449 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002450 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002451
2452 /* we necessarily have the response, so there are no partial failures */
2453 if (s->proxy->options2 & PR_O2_EXP_INV)
2454 ret = !ret;
2455
Simon Horman4a741432013-02-23 15:35:38 +09002456 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002457 break;
2458
2459 case PR_O2_EXP_STR:
2460 case PR_O2_EXP_RSTR:
2461 /* very simple response parser: ignore CR and only count consecutive LFs,
2462 * stop with contentptr pointing to first char after the double CRLF or
2463 * to '\0' if crlf < 2.
2464 */
2465 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002466 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002467 if (crlf >= 2)
2468 break;
2469 if (*contentptr == '\r')
2470 continue;
2471 else if (*contentptr == '\n')
2472 crlf++;
2473 else
2474 crlf = 0;
2475 }
2476
2477 /* Check that response contains a body... */
2478 if (crlf < 2) {
2479 if (!done)
2480 return 0;
2481
Simon Horman4a741432013-02-23 15:35:38 +09002482 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002483 "HTTP content check could not find a response body");
2484 return 1;
2485 }
2486
2487 /* Check that response body is not empty... */
2488 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002489 if (!done)
2490 return 0;
2491
Simon Horman4a741432013-02-23 15:35:38 +09002492 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002493 "HTTP content check found empty response body");
2494 return 1;
2495 }
2496
2497 /* Check the response content against the supplied string
2498 * or regex... */
2499 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2500 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2501 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002502 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002503
2504 /* if we don't match, we may need to wait more */
2505 if (!ret && !done)
2506 return 0;
2507
2508 if (ret) {
2509 /* content matched */
2510 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002511 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002512 "HTTP check matched unwanted content");
2513 else
Simon Horman4a741432013-02-23 15:35:38 +09002514 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002515 "HTTP content check matched");
2516 }
2517 else {
2518 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002519 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002520 "HTTP check did not match unwanted content");
2521 else
Simon Horman4a741432013-02-23 15:35:38 +09002522 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002523 "HTTP content check did not match");
2524 }
2525 break;
2526 }
2527 return 1;
2528}
2529
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002530/*
2531 * return the id of a step in a send/expect session
2532 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002533static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002534{
2535 struct tcpcheck_rule *cur = NULL, *next = NULL;
2536 int i = 0;
2537
Willy Tarreau213c6782014-10-02 14:51:02 +02002538 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002539 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002540 return 0;
2541
Simon Hormane16c1b32015-01-30 11:22:57 +09002542 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002543
2544 /* no step => first step */
2545 if (cur == NULL)
2546 return 1;
2547
2548 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002549 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002550 if (next->list.p == &cur->list)
2551 break;
2552 ++i;
2553 }
2554
2555 return i;
2556}
2557
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002558/*
2559 * return the latest known comment before (including) the given stepid
2560 * returns NULL if no comment found
2561 */
2562static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2563{
2564 struct tcpcheck_rule *cur = NULL;
2565 char *ret = NULL;
2566 int i = 0;
2567
2568 /* not even started anything yet, return latest comment found before any action */
2569 if (!check->current_step) {
2570 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2571 if (cur->action == TCPCHK_ACT_COMMENT)
2572 ret = cur->comment;
2573 else
2574 goto return_comment;
2575 }
2576 }
2577
2578 i = 1;
2579 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2580 if (cur->comment)
2581 ret = cur->comment;
2582
2583 if (i >= stepid)
2584 goto return_comment;
2585
2586 ++i;
2587 }
2588
2589 return_comment:
2590 return ret;
2591}
2592
Willy Tarreaube74b882017-10-04 16:22:49 +02002593/* proceed with next steps for the TCP checks <check>. Note that this is called
2594 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002595 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2596 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002597 *
2598 * Please do NOT place any return statement in this function and only leave
2599 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002600 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002601static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002602{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002603 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002604 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002605 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002606 struct conn_stream *cs = check->cs;
2607 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002608 struct server *s = check->server;
2609 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002610 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002611 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002612
Willy Tarreauef953952014-10-02 14:30:14 +02002613 /* here, we know that the check is complete or that it failed */
2614 if (check->result != CHK_RES_UNKNOWN)
2615 goto out_end_tcpcheck;
2616
2617 /* We have 4 possibilities here :
2618 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002619 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002620 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2621 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002622 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002623 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002624 * be waiting for a connection attempt to complete. conn!=NULL.
2625 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002626 *
2627 * #2 and #3 are quite similar, we want both the connection and the
2628 * handshake to complete before going any further. Thus we must always
2629 * wait for a connection to complete unless we're before and existing
2630 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002631 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002632
2633 /* find first rule and skip comments */
2634 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2635 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2636 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2637
Willy Tarreau00149122017-10-04 18:05:01 +02002638 if ((check->current_step || &next->list == head) &&
2639 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002640 /* we allow up to min(inter, timeout.connect) for a connection
2641 * to establish but only when timeout.check is set
2642 * as it may be to short for a full check otherwise
2643 */
2644 while (tick_is_expired(t->expire, now_ms)) {
2645 int t_con;
2646
2647 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2648 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2649
2650 if (s->proxy->timeout.check)
2651 t->expire = tick_first(t->expire, t_con);
2652 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002653 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002654 }
2655
Willy Tarreauef953952014-10-02 14:30:14 +02002656 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002657 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002658 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002659 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002660 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002661
Willy Tarreau213c6782014-10-02 14:51:02 +02002662 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002663 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002664 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002665 b_reset(&check->bo);
2666 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002667 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002668 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2669 if (s->proxy->timeout.check)
2670 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
2671 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002672
Willy Tarreauabca5b62013-12-06 14:19:25 +01002673 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002674 /* We have to try to flush the output buffer before reading, at
2675 * the end, or if we're about to send a string that does not fit
2676 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002677 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002678 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002679 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002680 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002681 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002682 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002683 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002684
Olivier Houcharded0f2072018-08-16 15:41:52 +02002685 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002686 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002687
2688 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002689 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002690 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002691 goto out_end_tcpcheck;
2692 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002693 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002694 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002695 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002696 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002697 goto out;
2698 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002699 }
2700
Willy Tarreau263013d2015-05-13 11:59:14 +02002701 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002702 break;
2703
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002704 /* have 'next' point to the next rule or NULL if we're on the
2705 * last one, connect() needs this.
2706 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002707 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002708
2709 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002710 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002711 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002712
2713 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002714 if (&next->list == head)
2715 next = NULL;
2716
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002717 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2718 struct protocol *proto;
2719 struct xprt_ops *xprt;
2720
Willy Tarreau00149122017-10-04 18:05:01 +02002721 /* For a connect action we'll create a new connection.
2722 * We may also have to kill a previous one. But we don't
2723 * want to leave *without* a connection if we came here
2724 * from the connection layer, hence with a connection.
2725 * Thus we'll proceed in the following order :
2726 * 1: close but not release previous connection
2727 * 2: try to get a new connection
2728 * 3: release and replace the old one on success
2729 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002730 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002731 cs_close(check->cs);
2732 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002733 }
2734
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002735 /* mark the step as started */
2736 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002737
2738 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002739 cs = cs_new(NULL);
2740 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002741 step = tcpcheck_get_step_id(check);
2742 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2743 comment = tcpcheck_get_step_comment(check, step);
2744 if (comment)
2745 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002746 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2747 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002748 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002749 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002750 }
2751
Willy Tarreauac59f362017-10-08 11:10:19 +02002752 if (check->cs)
2753 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002754
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002755 check->cs = cs;
2756 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002757 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002758 check->wait_list.events = 0;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002759 conn->target = &s->obj_type;
2760
2761 /* no client address */
2762 clear_addr(&conn->addr.from);
2763
Simon Horman41f58762015-01-30 11:22:56 +09002764 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002765 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002766 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002767 }
2768 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002769 /* we'll connect to the addr on the server */
2770 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002771 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002772 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002773
2774 /* port */
2775 if (check->current_step->port)
2776 set_host_port(&conn->addr.to, check->current_step->port);
2777 else if (check->port)
2778 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002779 else if (s->svc_port)
2780 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002781
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002782 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002783 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002784 }
2785 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002786 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002787 }
Willy Tarreaube373152018-09-06 11:45:30 +02002788
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002789 conn_prepare(conn, proto, xprt);
Olivier Houchardf502aca2018-12-14 19:42:40 +01002790 conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL);
Willy Tarreaube373152018-09-06 11:45:30 +02002791 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002792
Willy Tarreaue7dff022015-04-03 01:14:29 +02002793 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002794 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002795 ret = proto->connect(conn,
2796 1 /* I/O polling is always needed */,
2797 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002798 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2799 conn->send_proxy_ofs = 1;
2800 conn->flags |= CO_FL_SEND_PROXY;
2801 }
2802
2803 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002804 * - SF_ERR_NONE if everything's OK
2805 * - SF_ERR_SRVTO if there are no more servers
2806 * - SF_ERR_SRVCL if the connection was refused by the server
2807 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2808 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2809 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002810 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002811 * Note that we try to prevent the network stack from sending the ACK during the
2812 * connect() when a pure TCP check is used (without PROXY protocol).
2813 */
2814 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002815 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002816 /* we allow up to min(inter, timeout.connect) for a connection
2817 * to establish but only when timeout.check is set
2818 * as it may be to short for a full check otherwise
2819 */
2820 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2821
2822 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2823 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2824 t->expire = tick_first(t->expire, t_con);
2825 }
2826 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002827 case SF_ERR_SRVTO: /* ETIMEDOUT */
2828 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002829 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002830 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002831 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002832 comment = tcpcheck_get_step_comment(check, step);
2833 if (comment)
2834 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002835 set_server_check_status(check, HCHK_STATUS_L4CON,
2836 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002837 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002838 case SF_ERR_PRXCOND:
2839 case SF_ERR_RESOURCE:
2840 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002841 step = tcpcheck_get_step_id(check);
2842 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002843 comment = tcpcheck_get_step_comment(check, step);
2844 if (comment)
2845 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002846 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2847 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002848 goto out_end_tcpcheck;
2849 }
2850
2851 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002852 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002853
2854 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002855 while (&check->current_step->list != head &&
2856 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002857 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002858
Willy Tarreauf2c87352015-05-13 12:08:21 +02002859 if (&check->current_step->list == head)
2860 break;
2861
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002862 } /* end 'connect' */
2863 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2864 /* mark the step as started */
2865 check->last_started_step = check->current_step;
2866
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002867 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002868 if (*b_head(&check->bi) != '\0') {
2869 *b_head(&check->bi) = '\0';
2870 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002871 }
2872
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002873 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002874 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002875 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002876 goto out_end_tcpcheck;
2877 }
2878
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002879 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002880 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002881 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002882 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002883 set_server_check_status(check, HCHK_STATUS_L7RSP,
2884 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002885 goto out_end_tcpcheck;
2886 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002887
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002888 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002889 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002890 continue;
2891
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002892 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2893 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002894
Willy Tarreauabca5b62013-12-06 14:19:25 +01002895 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002896 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002897
2898 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002899 while (&check->current_step->list != head &&
2900 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002901 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002902
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002903 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002904 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002905 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002906 goto out_end_tcpcheck;
2907
Olivier Houchard511efea2018-08-16 15:30:32 +02002908 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002909 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002910 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002911 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002912 /* Report network errors only if we got no other data. Otherwise
2913 * we'll let the upper layers decide whether the response is OK
2914 * or not. It is very common that an RST sent by the server is
2915 * reported as an error just after the last data chunk.
2916 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002917 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002918 goto out_end_tcpcheck;
2919 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002920 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002921 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002922 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02002923 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002924 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002925 }
2926
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002927 /* mark the step as started */
2928 check->last_started_step = check->current_step;
2929
2930
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002931 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002932 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002933 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002934 if (b_data(&check->bi) < b_size(&check->bi)) {
2935 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002936 }
2937 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002938 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002939 done = 1; /* buffer full, don't wait for more data */
2940 }
2941
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002942 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002943
2944 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002945 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002946 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002947 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002948
2949 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002950 step = tcpcheck_get_step_id(check);
2951 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002952 comment = tcpcheck_get_step_comment(check, step);
2953 if (comment)
2954 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002955 set_server_check_status(check, HCHK_STATUS_L7RSP,
2956 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002957
2958 goto out_end_tcpcheck;
2959 }
2960
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002961 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002962 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002963
Willy Tarreaua970c282013-12-06 12:47:19 +01002964 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002965 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002966 ret = my_memmem(contentptr, b_data(&check->bi), check->current_step->string, check->current_step->string_len) != NULL;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002967 else if (check->current_step->expect_regex != NULL)
2968 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002969
2970 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002971 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002972
2973 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002974 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002975 if (ret) {
2976 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002977 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002978 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002979 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002980 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002981 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002982 }
2983 else {
2984 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002985 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002986 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002987 comment = tcpcheck_get_step_comment(check, step);
2988 if (comment)
2989 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002990 set_server_check_status(check, HCHK_STATUS_L7RSP,
2991 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002992 goto out_end_tcpcheck;
2993 }
2994 /* matched and was supposed to => OK, next step */
2995 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002996 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002997 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002998
2999 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003000 while (&check->current_step->list != head &&
3001 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003002 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003003
Willy Tarreauf2c87352015-05-13 12:08:21 +02003004 if (&check->current_step->list == head)
3005 break;
3006
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003007 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003008 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003009 }
3010 }
3011 else {
3012 /* not matched */
3013 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003014 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003015 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003016 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003017
3018 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003019 while (&check->current_step->list != head &&
3020 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003021 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003022
Willy Tarreauf2c87352015-05-13 12:08:21 +02003023 if (&check->current_step->list == head)
3024 break;
3025
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003026 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003027 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003028 }
3029 /* not matched but was supposed to => ERROR */
3030 else {
3031 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003032 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003033 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003034 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003035 }
3036 else {
3037 /* we were looking for a regex */
3038 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003039 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003040 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003041 comment = tcpcheck_get_step_comment(check, step);
3042 if (comment)
3043 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003044 set_server_check_status(check, HCHK_STATUS_L7RSP,
3045 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003046 goto out_end_tcpcheck;
3047 }
3048 }
3049 } /* end expect */
3050 } /* end loop over double chained step list */
3051
Baptiste Assmann248f1172018-03-01 21:49:01 +01003052 /* don't do anything until the connection is established */
3053 if (!(conn->flags & CO_FL_CONNECTED)) {
3054 /* update expire time, should be done by process_chk */
3055 /* we allow up to min(inter, timeout.connect) for a connection
3056 * to establish but only when timeout.check is set
3057 * as it may be to short for a full check otherwise
3058 */
3059 while (tick_is_expired(t->expire, now_ms)) {
3060 int t_con;
3061
3062 t_con = tick_add(t->expire, s->proxy->timeout.connect);
3063 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3064
3065 if (s->proxy->timeout.check)
3066 t->expire = tick_first(t->expire, t_con);
3067 }
3068 goto out;
3069 }
3070
Willy Tarreau263013d2015-05-13 11:59:14 +02003071 /* We're waiting for some I/O to complete, we've reached the end of the
3072 * rules, or both. Do what we have to do, otherwise we're done.
3073 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003074 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003075 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3076 goto out_end_tcpcheck;
3077 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003078
Willy Tarreau53c5a042015-05-13 11:38:17 +02003079 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003080 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003081 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003082 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003083
3084 out_end_tcpcheck:
3085 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003086 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003087 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003088
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003089 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003090 check->current_step = NULL;
3091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003092 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003093 conn->flags |= CO_FL_ERROR;
3094
Christopher Fauletb6102852017-11-28 10:06:29 +01003095 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003096 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003097}
3098
Simon Hormanb1900d52015-01-30 11:22:54 +09003099const char *init_check(struct check *check, int type)
3100{
3101 check->type = type;
3102
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003103 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3104 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003105
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003106 check->bi.area = calloc(check->bi.size, sizeof(char));
3107 check->bo.area = calloc(check->bo.size, sizeof(char));
3108
3109 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003110 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003111
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003112 check->wait_list.task = tasklet_new();
3113 if (!check->wait_list.task)
3114 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003115 check->wait_list.events = 0;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003116 check->wait_list.task->process = event_srv_chk_io;
3117 check->wait_list.task->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003118 return NULL;
3119}
3120
Simon Hormanbfb5d332015-01-30 11:22:55 +09003121void free_check(struct check *check)
3122{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003123 free(check->bi.area);
3124 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003125 if (check->cs) {
3126 free(check->cs->conn);
3127 check->cs->conn = NULL;
3128 cs_free(check->cs);
3129 check->cs = NULL;
3130 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003131}
3132
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003133void email_alert_free(struct email_alert *alert)
3134{
3135 struct tcpcheck_rule *rule, *back;
3136
3137 if (!alert)
3138 return;
3139
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003140 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3141 LIST_DEL(&rule->list);
3142 free(rule->comment);
3143 free(rule->string);
3144 if (rule->expect_regex)
3145 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003146 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003147 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003148 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003149}
3150
Olivier Houchard9f6af332018-05-25 14:04:04 +02003151static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003152{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003153 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003154 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003155 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003156
3157 q = container_of(check, typeof(*q), check);
3158
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003159 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003160 while (1) {
3161 if (!(check->state & CHK_ST_ENABLED)) {
3162 if (LIST_ISEMPTY(&q->email_alerts)) {
3163 /* All alerts processed, queue the task */
3164 t->expire = TICK_ETERNITY;
3165 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003166 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003167 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003168
3169 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003170 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003171 t->expire = now_ms;
3172 check->server = alert->srv;
3173 check->tcpcheck_rules = &alert->tcpcheck_rules;
PiBa-NL1714b9f2017-12-06 01:35:43 +01003174 check->status = HCHK_STATUS_UNKNOWN; // the UNKNOWN status is used to exit set_server_check_status(.) early
Christopher Faulet0108bb32017-10-20 21:34:32 +02003175 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003176 }
3177
Olivier Houchard9f6af332018-05-25 14:04:04 +02003178 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003179 if (check->state & CHK_ST_INPROGRESS)
3180 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003181
3182 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3183 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003184 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003185 check->server = NULL;
3186 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003187 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003188 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003189 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003190 return t;
3191}
3192
Christopher Faulet0108bb32017-10-20 21:34:32 +02003193/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3194 *
3195 * The function returns 1 in success case, otherwise, it returns 0 and err is
3196 * filled.
3197 */
3198int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003199{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003200 struct mailer *mailer;
3201 struct email_alertq *queues;
3202 const char *err_str;
3203 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003204
Christopher Faulet0108bb32017-10-20 21:34:32 +02003205 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3206 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003207 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003208 }
3209
Christopher Faulet0108bb32017-10-20 21:34:32 +02003210 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3211 struct email_alertq *q = &queues[i];
3212 struct check *check = &q->check;
3213 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003214
3215 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003216 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003217 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003218 check->rise = DEF_AGENT_RISETIME;
3219 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003220 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3221 memprintf(err, "%s", err_str);
3222 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003223 }
3224
3225 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003226 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003227 check->port = get_host_port(&mailer->addr);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003228 //check->server = s;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003229
Emeric Brunc60def82017-09-27 14:59:38 +02003230 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003231 memprintf(err, "out of memory while allocating mailer alerts task");
3232 goto error;
3233 }
3234
3235 check->task = t;
3236 t->process = process_email_alert;
3237 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003238
Christopher Faulet0108bb32017-10-20 21:34:32 +02003239 /* check this in one ms */
3240 t->expire = TICK_ETERNITY;
3241 check->start = now;
3242 task_queue(t);
3243 }
3244
3245 mls->users++;
3246 free(p->email_alert.mailers.name);
3247 p->email_alert.mailers.m = mls;
3248 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003249 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003250
3251 error:
3252 for (i = 0; i < mls->count; i++) {
3253 struct email_alertq *q = &queues[i];
3254 struct check *check = &q->check;
3255
3256 if (check->task) {
3257 task_delete(check->task);
3258 task_free(check->task);
3259 check->task = NULL;
3260 }
3261 free_check(check);
3262 }
3263 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003264 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003265 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003266}
3267
3268
3269static int add_tcpcheck_expect_str(struct list *list, const char *str)
3270{
3271 struct tcpcheck_rule *tcpcheck;
3272
Willy Tarreaubafbe012017-11-24 17:34:44 +01003273 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003274 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003275 memset(tcpcheck, 0, sizeof(*tcpcheck));
3276 tcpcheck->action = TCPCHK_ACT_EXPECT;
3277 tcpcheck->string = strdup(str);
3278 tcpcheck->expect_regex = NULL;
3279 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003280 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003281 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003282 return 0;
3283 }
3284
3285 LIST_ADDQ(list, &tcpcheck->list);
3286 return 1;
3287}
3288
3289static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3290{
3291 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003292 const char *in;
3293 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003294 int i;
3295
Willy Tarreaubafbe012017-11-24 17:34:44 +01003296 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003297 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003298 memset(tcpcheck, 0, sizeof(*tcpcheck));
3299 tcpcheck->action = TCPCHK_ACT_SEND;
3300 tcpcheck->expect_regex = NULL;
3301 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003302 tcpcheck->string_len = 0;
3303 for (i = 0; strs[i]; i++)
3304 tcpcheck->string_len += strlen(strs[i]);
3305
3306 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3307 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003308 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003309 return 0;
3310 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003311
Willy Tarreau64345aa2016-08-10 19:29:09 +02003312 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003313 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003314 for (in = strs[i]; (*dst = *in++); dst++);
3315 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003316
3317 LIST_ADDQ(list, &tcpcheck->list);
3318 return 1;
3319}
3320
Christopher Faulet0108bb32017-10-20 21:34:32 +02003321static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3322 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003323{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003324 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003325 struct tcpcheck_rule *tcpcheck;
3326 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003327
Willy Tarreaubafbe012017-11-24 17:34:44 +01003328 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003329 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003330 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003331 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003332 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003333
Willy Tarreaubafbe012017-11-24 17:34:44 +01003334 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003335 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003336 memset(tcpcheck, 0, sizeof(*tcpcheck));
3337 tcpcheck->action = TCPCHK_ACT_CONNECT;
3338 tcpcheck->comment = NULL;
3339 tcpcheck->string = NULL;
3340 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003341 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3342
3343 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3344 goto error;
3345
3346 {
3347 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3348 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3349 goto error;
3350 }
3351
3352 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3353 goto error;
3354
3355 {
3356 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3357 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3358 goto error;
3359 }
3360
3361 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3362 goto error;
3363
3364 {
3365 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3366 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3367 goto error;
3368 }
3369
3370 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3371 goto error;
3372
3373 {
3374 const char * const strs[2] = { "DATA\r\n" };
3375 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3376 goto error;
3377 }
3378
3379 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3380 goto error;
3381
3382 {
3383 struct tm tm;
3384 char datestr[48];
3385 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003386 "From: ", p->email_alert.from, "\r\n",
3387 "To: ", p->email_alert.to, "\r\n",
3388 "Date: ", datestr, "\r\n",
3389 "Subject: [HAproxy Alert] ", msg, "\r\n",
3390 "\r\n",
3391 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003392 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003393 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003394 NULL
3395 };
3396
3397 get_localtime(date.tv_sec, &tm);
3398
3399 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3400 goto error;
3401 }
3402
3403 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3404 goto error;
3405 }
3406
3407 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3408 goto error;
3409
3410 {
3411 const char * const strs[2] = { "QUIT\r\n" };
3412 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3413 goto error;
3414 }
3415
3416 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3417 goto error;
3418
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003419 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003420 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003421 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003422 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003423 return 1;
3424
3425error:
3426 email_alert_free(alert);
3427 return 0;
3428}
3429
Christopher Faulet0108bb32017-10-20 21:34:32 +02003430static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003431{
3432 int i;
3433 struct mailer *mailer;
3434
3435 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3436 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003437 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003438 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003439 return;
3440 }
3441 }
3442
3443 return;
3444}
3445
3446/*
3447 * Send email alert if configured.
3448 */
Simon Horman64e34162015-02-06 11:11:57 +09003449void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003450{
3451 va_list argp;
3452 char buf[1024];
3453 int len;
3454 struct proxy *p = s->proxy;
3455
Christopher Faulet0108bb32017-10-20 21:34:32 +02003456 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003457 return;
3458
3459 va_start(argp, format);
3460 len = vsnprintf(buf, sizeof(buf), format, argp);
3461 va_end(argp);
3462
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003463 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003464 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003465 return;
3466 }
3467
Christopher Faulet0108bb32017-10-20 21:34:32 +02003468 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003469}
3470
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003471/*
3472 * Return value:
3473 * the port to be used for the health check
3474 * 0 in case no port could be found for the check
3475 */
3476int srv_check_healthcheck_port(struct check *chk)
3477{
3478 int i = 0;
3479 struct server *srv = NULL;
3480
3481 srv = chk->server;
3482
3483 /* If neither a port nor an addr was specified and no check transport
3484 * layer is forced, then the transport layer used by the checks is the
3485 * same as for the production traffic. Otherwise we use raw_sock by
3486 * default, unless one is specified.
3487 */
3488 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003489 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003490 chk->send_proxy |= (srv->pp_opts);
3491 }
3492
3493 /* by default, we use the health check port ocnfigured */
3494 if (chk->port > 0)
3495 return chk->port;
3496
3497 /* try to get the port from check_core.addr if check.port not set */
3498 i = get_host_port(&chk->addr);
3499 if (i > 0)
3500 return i;
3501
3502 /* try to get the port from server address */
3503 /* prevent MAPPORTS from working at this point, since checks could
3504 * not be performed in such case (MAPPORTS impose a relative ports
3505 * based on live traffic)
3506 */
3507 if (srv->flags & SRV_F_MAPPORTS)
3508 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003509
3510 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003511 if (i > 0)
3512 return i;
3513
3514 return 0;
3515}
3516
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003517REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003518
Willy Tarreaubd741542010-03-16 18:46:54 +01003519/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003520 * Local variables:
3521 * c-indent-level: 8
3522 * c-basic-offset: 8
3523 * End:
3524 */